Entropy Analysis and Complex Patient-ventilator Interactions During Mechanical Ventilation

Overview

Complex patterns of patient-ventilator interactions could be miscalculated by visual observation of mechanical ventilator screens or current algorithms based on physiologic waveforms to detect patient-ventilator asynchronies. Therefore, we aim to characterize, validate and study the clinical distribution and implications of an automated and personalized non-invasive tool based on Entropy to detect Complex Patient-Ventilator Interactions (CP-VI) during mechanical ventilation, defined as breathing pattern change and/or clusters of asynchronies, over the signals of airway pressure (Paw) and airway flow (Flow).

Methods 1. Defining complex patient ventilator interactions The Investigators defined "Complex Patient-Ventilator Interactions" (CP-VI) as the presence alone or in combination of a change in the respiratory rate of more than 50%, and/or the occurrence of any kind of asynchronies 2. Data acquisition and data analysis Paw and Flow signals will continuously recorded throughout patient's stay in the Intensive Care Unit (UCI) using BetterCare® system (Better Care®, Barcelona, Spain). BetterCare uses drivers specifically designed to interact with the output signal of mechanical ventilators and bedside monitors rather than directly with patients. Recorded signals are synchronized and stored for further analysis. MATLAB (The MathWorks, Inc., vR2018b, Natick, MA, USA) will be used to perform the signal processing, data analysis and visual assessment. 3. Study Population The Investigators will obtain data from an prospectively constructed database from a connectivity platform (Better Care®) to interoperate signals from different ventilators and monitors and subsequently compute algorithms to diagnose patient-ventilator asynchronies (ClinicalTrial.gov, NCT03451461). All of those patients corresponding to a self-extubation event previously recorded will be recruited for the characterization and validation process, in order to guarantee at patient-ventilator interactions and episodes when they fight the ventilator. Also, patients in whom an spontaneous breathing trial previous to an attempt to librate him/her from the ventilator will be recruited in order to obtain signals of Paw and Flow. Clinical and demographic data will obtained from the medical chart. The institutional review board approved the protocol and waived informed consent because the study was non-interventional, posed no added risk to patients, and did not interfere with usual care. 4. Visual validation of CP-VI Three researchers will visually review the Flow and Paw recordings of events. The segments duration will selected based on previous studies where asynchronous clusters are evaluated. The dataset will be previously selected by an expert in mechanical ventilation ensuring balance by ventilation modes (grouped by Pressure Support Ventilation (PSV) and Assist-Control Ventilation modes) and equal distribution of CP-VI presence or absence. The controlled modes included volume assist-control ventilation (VACV) and pressure assist-control ventilation (PACV). Flow and Paw tracings will be randomly ordered in MATLAB prior to visual analysis to ensure blinding of the scorers. Scorers will be provided with written description of CP-VI characteristics before visual analysis, as a reference. On base of CP-VI definition previously described, each researcher will score for the presence or absence of CP-VI events, without time limitation. The visual assessment will considered as the gold standard. 5. Entropy Entropy is a non-linear measure that allows assessing the randomness of a series of data. Entropy calculation requires three parameters: the embedding dimension, m (a positive integer); the tolerance value or similarity criterion, r (a positive real number); and the total length, N, of the analysed series. 6. Automatic CP-VI detection An automated algorithm for CP-VI detection based on Entropy tool will be implemented. 7. Statistical analysis Fleiss's kappa coefficient will be used as reliability of agreement among raters for visual assessment. The automated algorithm for CP-VI detection will be applied over the entropy series derived from the same Flow and Paw. The performance of the automated algorithm will be evaluated on base of sensitivity (Se), specificity (Sp), positive and negative predictive values (PPV and NPV, respectively), accuracy (ACC) and Matthews correlation coefficient (MCC). 8. Selection of m, r and N In entropy studies, an important step is to determine the optimal settings to robust extract the randomness of a series of data. Therefore, an optimization procedure of m, r and N will performed to properly estimate CP-VI.